A reminder - you can access the full MAB20 proceedings on dl.acm.org/doi/proceedings/1…

A Dubai-based photographer spent 10 years searching for a black lion once considered a myth. Using drones and thermal imaging cameras. he covered tens of thousands of kilometers in Africa until. in 2025. sensors detected an elusive "shadow" in a reserve--the first confirmation of a melanistic lion in a century The "Black Ghost,' exiled from his pride because of his coloring, survived alone, hunting onlv at night. His uniaue mutation (one in 15 million individuals) makes daytime life impossible, but the lion nevertheless reached an impressive size--120 cm at the shoulder and over 220 ka in weiaht Genetics confirmed the authenticity of the mutation, and the lion's territory is now protected. The photographer released the images for free to show the world that natural wonders cannot be replicated even with modern technology.

Yann LeCun was right the entire time. And generative AI might be a dead end. For the last three years, the entire industry has been obsessed with building bigger LLMs. Trillions of parameters. Billions in compute. The theory was simple: if you make the model big enough, it will eventually understand how the world works. Yann LeCun said that was stupid. He argued that generative AI is fundamentally inefficient. When an AI predicts the next word, or generates the next pixel, it wastes massive amounts of compute on surface-level details. It memorizes patterns instead of learning the actual physics of reality. He proposed a different path: JEPA (Joint-Embedding Predictive Architecture). Instead of forcing the AI to paint the world pixel by pixel, JEPA forces it to predict abstract concepts. It predicts what happens next in a compressed "thought space." But for years, JEPA had a fatal flaw. It suffered from "representation collapse." Because the AI was allowed to simplify reality, it would cheat. It would simplify everything so much that a dog, a car, and a human all looked identical. It learned nothing. To fix it, engineers had to use insanely complex hacks, frozen encoders, and massive compute overheads. Until today. Researchers just dropped a paper called "LeWorldModel" (LeWM). They completely solved the collapse problem. They replaced the complex engineering hacks with a single, elegant mathematical regularizer. It forces the AI's internal "thoughts" into a perfect Gaussian distribution. The AI can no longer cheat. It is forced to understand the physical structure of reality to make its predictions. The results completely rewrite the economics of AI. LeWM didn't need a massive, centralized supercomputer. It has just 15 million parameters. It trains on a single, standard GPU in a few hours. Yet it plans 48x faster than massive foundation world models. It intrinsically understands physics. It instantly detects impossible events. We spent billions trying to force massive server farms to memorize the internet. Now, a tiny model running locally on a single graphics card is actually learning how the real world works.

new article just published: Beyond the Looking Glass: multimodal LLM depth-sensing ... frontiersin.org/journals/com…

published! the first article in the Research Topic: Re-Imagining Mediated Human Building Interaction and Sensory Environments: Volume II frontiersin.org/journals/com…

Meanwhile, in London.. an AI humanoid was seen walking the streets and was surrounding by intrigued school kids. Spooky.

It is that time of the year! #winterlight2026 canarywharf.com/winter-light…

After twelve years of work, the world’s most beautiful subway station has been inaugurated in Rome: Colosseo, an underground archaeological museum.

So, today @tedtalks announced that my TED talk ‘This is What a Digital Coup Looks Like’ was their most watched talk of the year. It’s sort of amazing & terrifying. Because this *is* what a digital coup looks like.

Life of a keyboard.

This is going to eliminate human Barbers 😂

📢 New paper in Nature Human Behavior! A perspective on the field we have been working in the last 10 years: urban mobility data reveals experienced inequalities in cities beyond where people live. 👉 Link: nature.com/articles/s41562-0… 👉 Open-access: rdcu.be/eabf4

Dear friends, today is the day! We are excited to share the Living Painting series of Biome Lumina. This V3 update allows our collectors to explore the MyDataland platform and have real-time interactions with their Biome. Sound, vision, and chat modules are ultra-inspiring! :)

ChatGPT is starting to look less like a chatbot and more like...an operating system wired.com/story/openai-dev-d… latest from me & @willknight

On #WorldArchitectureDay, we wanted to share some of our favourite photographs of the Abbey church with you, beginning with this spectacular view of the ceiling above the nave and transepts. The nave soars 101ft (31 metres) above the Abbey floor, making it the tallest Gothic nave in England.

OpenAI just launched its own version of an app store, taking aim at Apple and Google trib.al/S7FCuSq

🚨 BREAKING: The second major European open-source LLM is OUT, and it has been announced as multilingual and fully compliant with the EU AI Act. Is the EU finally catching up? Here's what you need to know: TildeOpen LLM is a 30-billion-parameter model optimized for European languages. The model was trained using the 2 million GPU hours awarded through the AI Grand Challenge on the EuroHPC LUMI supercomputer. According to the official announcement, the model's key features are: - State-of-the-art performance across all 24 EU official languages, plus Ukrainian, Norwegian, and several Balkan languages - Superior efficiency compared to leading global models (smaller, faster, and more accessible) - Full compliance with the EU AI Act - Data security maintained within the EU infrastructure - The EU is under extreme pressure to innovate and show that it can be a competitive player in the AI race (especially after the negative feedback from the Draghi report). It's nice to see the EU fostering internal initiatives that rely on EU funds and EU infrastructure, and that follow the "European way," prioritizing openness, multilingualism, and legal compliance. By the way, if the goal is to innovate, strengthen the internal AI infrastructure, and become more competitive in AI, initiatives like this one (and the Swiss AI model 'Apertus') make MUCH MORE sense than simplifying the GDPR and postponing the EU AI Act... - 👉 Never miss my updates, curations, and analyses on AI: join my newsletter's 79,800 subscribers (link below).

Oh god my brain just switched to a horrifying dark mirror universe where next election it's AI Winston Churchill vs AI Tony Blair.

RIBA unveils rebrand with "bold new logotype" by Johnson Banks: dezeen.com/2025/10/02/riba-r…